Capabilities Of Distributed Eeg Data Processing Computer Science Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

1Control System and Signal Laboratory, Dept. of Electrical Engineering, Technological Institute of Patras, Patras 24334, Greece

In this work a remote controlled system for acquiring and processing EEG signals is presented having also capabilities of distributed EEG data processing. The system consists of a PC running a specialized software and a specialized data acquisition (DAQ) card. The software environment for contacting remote EEG signal acquisition, analysis and data access is based on the well known National Instrument�s LabVIEW. EEG data can be transmitted over the internet and the associated software provides security in EEG and diagnostic analysis data sharing preventing access by unauthorized viewers. The EEG data processing is based on Independent Component Analysis technique (ICA). This technique allows good separation of signals arising from brain and non-brain sources (artifacts), as well as the distinction of the temporal discrete but spatial overlapping brain activities. The presented system is used for sharing confidential EEG data over the internet providing an automatic decision-making tool for a �second opinion� analysis in a clinical diagnosis procedure.

Index Terms: Distributed processing, EEG, ICA, Remote Monitoring, Telemedicine.


The demand for remote EEG data transmission mainly comes from doctor specialists� great difficulty in accessing underserved and isolated communities such as islands or mountain villages. In addition, development of such systems is motivated by the disability of elderly people to access clinics and hospitals, and by the need for decreasing the costs of unnecessary patient transfers from the towns to hospitals in the capital cities [1]. To this end many solutions have been developed for data transferring [2]. Most of them are using Application Data Interfaces � APIs, which they simplify the programming procedure of the communication part while they provide security in data sharing [3]. At the same time the introduction of the innovative ICA method [4], come up with very significant results in EEG analysis[5][6]. Using this method, systems able to acquire, transfer and process data (in real time or not) have been developed [7]. To this context, a system that provides distributed processing, allowing access of the recorded EEG data as well as the analysis results to a number of subscribers is very important. This sharing feature could play an important role especially in diagnostic procedures, when a second opinion is needed. Second opinion could be produced by inspecting EEG data and analysis results from a different view point a process that can be facilitated by using a distributed diagnosis system.

To this context in this work is described an �intelligent� EEG recording and analysis system which combines the advantages of the remote data access and control of an EEG DAQ system. The systems employs EEG analysis features by using the innovative ICA method forming a complete and effective solution.


3.1. System Architecture.

The general topology of the proposed system is depicted in Fig. 1. The philosophy of the system is based on distant EEG data access and the distribution of the execution control via Data Socket. Additionally every client must have the ability of unique data processing, while the results should be shared among many subscribers. In this way, a variety in data analysis can be achieved. Moreover every client must have the capability of making adjustments about the acquisition process via a messenger task which is also included in the system software.

Using DataSockets we succeed to decrease the developing time of the communication part of our application, to achieve data transferring and processing in real time, to ensure high level data protection, decreasing the total processing time due to the distribution of the various processes, and finally a full duplex communication between subscribers.

Fig 1: The Architecture of our system.

The Server which is a PC based DAQ system, is located in the room where the experiment takes place. First of all it acquires transmits and processes the EEG data. At the end of the experiment applies ICA algorithms to the recorded data, and shares the resulted unmixing matrix that ICA algorithms return to all of the subscribers. At the same time the DAQ system receives processed results from subscribers. Also all the subscribers can exchange ICA algorithms results between each other.

3. 2. Data acquisition and transfer technology.

Our data acquisition and transferring can be separated in hardware, and software components.

3. 2. 1. Hardware.

For acquiring the signals we used four electrodes with a screened cable. We designed a signal conditioning card, employing a low-pass filter with a cut off frequency of 49 Hz and a common mode rejection unit cancelling electrodes noise. The output of the conditioning card is connected to a NI-DAQ-9215A data acquisition card.

3. 2. 2. Software.

We produce a unique application code for every client workstation. This application provides the presentation, communication and data processing with other clients and with the server. The software is make use of LabVIEW ability to include MATLAB codes by incorporating algorithms for performing ICA written in MATLAB language. Every subscriber executes one different ICA algorithm form a total of four algorithms which are implemented. More specifically the server executes Runica which is an automated version of the extended Infomax algorithm [8], the client No1 executes the JadeR algorithm [9], the client No2 executes FastICA [10] and finally the Client No3 executes efICA [11] algorithm.

3. 2. 3. TCP-IP protocol in cooperation with the Application Programming Interface (API) LabVIEW DataSocket.

The use of the Internet for the remote monitoring of the patients has significantly increased over the last years. The most important feature of TCP-IP protocol is that it is able to provide real-time data transferring. It can be used in many applications like distributed DAQ systems, remote monitoring and control.


Independent component Analysis [4], was first developed in order to solve the blind source separation problem. More specifically, for a given set of mixed signals, promises to recover the source signals based on independence. The mixed signals in the basic model, are supposed to constitute a linear combination of the source signals with the coefficients given by a mixing matrix . Consequently the basic ICA model is described by the equation,

The ICA method seeks to find a unmixing matrix W in order to reverse the result of the mixing process. As an output of this process we receive n independent components, , which they are described by the vector-matrix notation .

The theoretical approach of ICA is based on statistical independence of the sources. As it is stated by the central limit theorem, mixed signals tend to follow the Gaussian distribution more strongly than the source signals. Therefore ICA is based in high order statistics (eg. Kurtosis), in contradiction with other similar methods (eg. PCA) which is based on second order statistics. ICA return a new set of data which are uncorrelated because independence implies uncorrelatedness but the reverse doesn�t hold. That�s why Independence is a stronger criterion than uncorrelatedness. Statistical independence can be expressed in terms of probabilities by the following equation,

which means that the joint probability density function (pdf) is equal with the product of the marginal pdf �s.

Many algorithms have been developed to perform ICA. However it has been proved by many authors that they are based on the same criterion followed with a different approach. Some of the most commonly used algorithms that are used in our data analysis are, �Infomax� [8] (in the main workstation we use an automated version of this algorithm called Runica), �FastICA�, �JadeR� and �efICA�.

4. 1. ICA for EEG data analysis.

ICA when used for EEG data analysis imposes two restrictions [12].

1. The source signals are constituted by linear mixtures of temporal independent but spatial fixed brain and no-brain activities,

2. The spatial spread of electric current from sources by volume conduction does not involve significant time delays.

In single trial EEG analysis, the rows of the matrix x are signals recorded from different electrodes and the columns are measurements recorded at different time instants. Performing ICA to the set of �mixed� signals x, we receive an unmixing matrix W who linearly unmixes the mixed signals into temporal independent and spatial fixed components, following the relation y=Wx. The rows of the output matrix y are time courses of activation of independent components. The projection of these Independent Components at each scalp sensor is given by the columns of the inverse of the unmixing matrix In this way we can extract the scalp topography of each component and also its physiological origins. In this way we can define artifacts [13], [14], which in turn can be easily removed from the new data set that ICA gives us.


The experiment data collection procedure took place in the laboratory of the Signal Processing and Systems A.T.E.I. Patras. Continuous EEG signal acquiring has been achieved using the PC-based DAQ system. This is realized by connecting four electrodes on a volunteer�s head, in distances based on 10-20 system for a time of ten minutes. The data are recorded, presented and analyzed locally in the PC-based DAQ system and also on three long distance workstations. Every workstation could define its own processing parameters and has the ability to exchange data with other workstations including the acquisition workstation. The main workstation where the DataSocket server runs is the PC that acquires the data from the DAQ device. In Fig. 2 a sample of 4 second recording from the server front panel is depicted. The sampling frequency has been defined to be 512 Hz, and a has been employed a digital Bandpass adjustable filter with low cutoff frequency at 0.1 Hz, and high cutoff frequency at 48Hz.

Fig. 2. Main workstation front panel.

Fig. 3. Client no 1

5. 1. Data processing and experiment results

In this section are presented data that they were processed using the EEGLAB [15] functions implemented in MATLAB and nested in LabVIEW. Our intent is to recover the independent components by performing ICA on the recorded data in order to discriminate brain and non brain activities [16] [17].

After a ten minutes continuous recording of EEG signals the data were transferred through the network to the subscribers. At the end of the recording procedure all of the subscribers perform ICA analysis and exchange their results.

In the server the Infomax algorithm is employed by using the EEGLAB automated version of the algorithm named Runica. The rest three work stations performed FastICA JadeR and Efficient ICA respectively. As preprocessing, in all the workstations was applied an IIR Bandpass filter with band pass (0.1- 48Hz). In Fig. 4, five seconds of the recorded EEG signals is presented.

Fig. 4. Data scroll activations.

In Fig. 5, the channel spectra and maps of the recorded signals is shown.

Fig. 5. Channel spectra and maps.

After applying the algorithm in the recorded data at every workstation we received four temporal Independent and four spatially fixed Components. In the next figure we present the results of applying runica() algorithm on the main workstation. The results are represented using EEGLAB functions. In the following figure we show the temporal independent activations and the respective spatially fixed components.

Fig. 3. Temporal independent and spatially fixed Components.

In these results is a noticeable symmetry on the ICA activity distribution over the scalp. The power spectra and maps of the Independent Components are shown in figure 6.

Fig. 6. Power spectra and maps of the ICs.

The DFT of the ICs is presented in the following figure.

Fig. 7. The DFT of the ICs.

From the last figure is clearly defined that the line noise (50Hz) is still present in the filtered data.


The implementation of such a distributed system is expected to be followed by an important development in a wide field of applications. We mention some of them:

1. Signal Processing and data representation.

Although the fast development of this science branch, it is steal needed to develop new more efficient or improving the previous techniques.

2. Real time applications.

Data are ideally transferred as fast as possible achieving real time communication. By developing more efficient protocols we could succeed faster data transferring that provide a more ideally real time communication.

3. Developing algorithms for clinical decision making tools.

It is needed to develop algorithms that have the ability to implement a typical diagnosis. By this way it could be possible, to detect abnormalities during the examination, the timely notification in critical situations like epileptic crisis.

4. Telemedicine.

The combination of the above views resulted in the development of completed telemedicine applications. For instance a patient with movement disabilities could be monitoring by a specialist while be at his home. In an emergency situation, where that system detects an abnormality, it could timely give an alarm to a specific person.